Friday, July 06, 2007

Delinquencies and Defaults for UberNerds

by Tanta on 7/06/2007 04:20:00 PM

I have no intention of exhausting the topics of delinquency and default today. My goal is rather more modest than that; I merely want to introduce to the non-mortgage-backed-security world a few definitions of terms, in hopes that perhaps some of these startling numbers being thrown around in the press can be interpreted. Besides that, it's too early in the day to start drinking, so we might as well waste our time being educated.

The fact is we see a lot of press reports quoting delinquency rates as if they were default rates. A large part of the problem, besides the general lack of expertise of a lot of business reporters, is that generally the only freely-available numbers to report are delinquency rates expressed as a percentage of a given book of business. There are other, more sophisticated ways of measuring credit loss risk than a simple bucketing of loans into categories of "current" and "delinquent," but these specialist calculations are generally not available to those who do not pay for subscriptions to analytic services. One of the most important calculations is the "roll rate," or the rate at which current or delinquent loans are "rolling" into the next bucket (current to 30 days, 30 days to 60 days, etc.). Another hugely important measure is the default rate: the projection of actual defaults for a given bucket of current or delinquent loans.

It is not possible, of course, to understand default rates without understanding the distinction between "delinquency" and "default." There are alternative ways of measuring delinquency (the "OTS method" and the "MBA method"), but it comes down to the same thing: a delinquent loan has missed at least one scheduled payment. Therefore, a "30-day" delinquent loan is past due by one payment as of the report date; a "60-day" is past due by two payments, and so on. The trouble is that these may or may not be consecutive or even most recent payments.

30-day delinquencies are very volatile. They are often seasonal, for one thing, and they can very often turn into what underwriters call a "rolling 30" ("rolling" in this case is not to be confused with a "roll rate" of 30 to 60 days delinquent). Imagine a borrower who is current through the June payment, misses the July payment, makes the August payment, and continues on through the end of the year making a payment each month, but never making up that missed payment from July. That borrower would be considered "current" in July (you aren't "delinquent" for our purposes until you're at least 30 days delinquent), 30 days deliquent in August, 30 days delinquent in September, and so on to the end of the year. It doesn't matter how far in time you get from that missed payment: you only missed one, so you are never on any given reporting date more than 30 days down.

You can have "rolling 60s," but servicers are rather less likely to put up with them than rolling 30s. At some point a servicer has to decide whether to "accelerate" the loan or not, and that's always a tough call with a rolling delinquency, since the very fact that it's "rolling" means that the borrower has resumed making payments; the borrower simply has not caught up with the misssed payment or payments. These borrowers are generally the best candidates for a modification. The lender can add the missed payment to the loan balance, which brings it "current" again and stops the accumulation of late fees and endless collection calls, among other things.

Things like these "rolling 30s" do make it unwise to use simple metrics like a 30-day delinquency rate (the percentage of loans by balance or units in a given pool or portfolio that are 30 days delinquent as of the report date) to predict future losses, since a simple delinquency rate calculation cannot distinguish between a "new 30" and a "rolling 30," and loans that are 30-days delinquent last month can certainly become current this month, as borrowers do sometimes make up missed payments. This is the main reason that most reports distinguish between "delinquency" and "serious delinquency," the latter generally defined as 60+ or 90+ lates.

As a general rule, most servicers do not begin foreclosure proceedings until a loan is 90 days delinquent. This isn't a "magic number," it's just a rule of thumb. For our purposes, one thing it means is that a servicer will usually report separately on loans "90 days delinquent" and "loans in foreclosure." "In foreclosure" means the loan is in a process, often a long one, that starts with the filing of a foreclosure notice and ends at an auction on the courthouse steps. Because there can be loans that are 90 days or more delinquent that are not in foreclosure--there could be forbearance or short sale or other workout arrangements in process, or a bankruptcy stay could be delaying the foreclosure filing--it is important to verify, with any set of numbers you look at, whether a "delinquent" bucket includes loans in foreclosure or not, to avoid double-counting when the numbers are totaled.

It makes a big difference, because the "roll rate" or likelihood of eventual default of a 90-day delinquent loan is not the same as the "roll rate" of a loan in foreclosure. Two such loans might actually be past-due for the same amount of time, but the fact that the servicer initiated foreclosure proceedings in one case but not the other is, precisely, a judgment of relative likelihood of default. Plenty of people have plenty of suspicions about how wise some of these servicer judgments are--that's the whole uproar over modifications and forbearance in a nutshell--but the point is that once a servicer files a foreclosure notice and begins the process, ultimate default of the loan is more likely just because the process is in motion.

This is probably the point where we should define "default." That term has an everyday legal definition that may confuse people in the context of credit losses on mortgage securities. In everyday terms any violation of a contract, including a mortgage, is a "default." You can, technically, be "in default" of your mortgage by failing to pay a $17.50 HOA assessment or moving out of your home, even if you keep your mortgage payments current. For purposes of calculating loss probability on a pool of mortgage loans, "default" means something much more specific: a defaulted loan, according to the Bond Market Association, "no longer pays principal and interest and then remains delinquent until liquidated." There are many ways a loan can liquidate: it can be foreclosed and the property purchased by a third party; it can become REO and the REO liquidated by sale to a third party; it can be a short sale or short payoff (the latter being a refinance of a delinquent loan in which the noteholder accepts less than full payoff).

You have to remember that "default" does not mean "loss." Imagine a $100,000 loan which becomes delinquent, is foreclosed, becomes REO, and is ultimately liquidated by sale of the REO. This loan is defaulted, but the loss to the noteholder might be only $20,000 after application of sale proceeds. That would mean that $80,000 comes back to the noteholder in the form of principal repayment. The importance of measuring defaults accurately is that it allows the security accounting to distinguish between normal or voluntary prepayments (principal returned via a performing loan's refinance or sale of the subject property) and involuntary prepayments via recovery from liquidation of delinquent loans. But the purpose of reporting default balances is not to equate defaults with net losses. The latter will depend on "loss severity," or what percentage of the outstanding loan balance is not recovered in liquidation. That, obviously, is an issue of home prices and servicer efficiency.

The problem, then, is getting from delinquencies to defaults, in terms of projecting what ultimate losses will be. Here's an example, from Fitch, of how one might project losses on a seasoned mortgage pool:

The first column puts the loan pool in "buckets" according to delinquency status as of 18 months age of the pool. The second column projects lifetime defaults for each bucket, using a series of assumptions and calculations based on historical experience, performance of this specific pool in its first 18 months compared to initial projections, home price appreciation estimates, and so on. The projected defaults range from 11% for "current" loans to 76% for "foreclosure" loans (the REO category is 100%, since once foreclosure is completed there is no possiblity of a loan becoming current again, and the definition of default is the liquidation of a delinquent loan).

The third column is important, because while "current" loans have the lowest likelihood of default (11%), they are the largest bucket in the pool, and so "current" loans that ultimately default are projected to be 9.1% of the pool, as opposed to "foreclosure" loans with the highest likelihood of default (76%), which are projected to be 2.9% of the pool.

The fourth column is the loss severity, or the percentage of the loan balance that is not expected to be recovered after liquidation. For the purposes of a table like this, remember that it is an average: some loans may have much higher recoveries, and some much lower. Obviously it is a crucial calculation for determining ultimate losses.

The final column gives you expected loss as a percent of the current pool balance, which in this example is 63% of the original balance. (The original balance has been reduced by 37% prepayments.) The projected loss as a percentage of original pool balance is 5.25%, or 7.1% of current pool balance.

Note that the pool has already experienced losses of 0.77% in its first 18 months. That does not necessarily mean that the lowest-rated tranche has taken a write-down; in fact, overcollateralization and excess spread on most deals should be more than sufficient to cover losses of 0.77% in 18 months without "breaking into" the principal balance of the lowest-rated tranche. The significance of the projections you see on this chart are, for an actual bond, a question of how the timing of projected losses play out as well as the adequacy of the OC and excess spread. It is possible that this pool will never see actual principal write-downs, even if it achieves losses as projected in the chart, if its OC and excess are adequate. (They may well not be adequate, but this chart doesn't tell you that one way or the other.)

It should be clear, but let me be tedious: the projections of loss in the example here are over the remaining life of the pool, not over the next year or month or week. Such projections may or may not be accurate: projections are projections, not divinations. I bring this example up not to suggest that it produces a loss prediction we can use as a rule of thumb: this is a specific example pool and your mileage may vary on a different pool. My purpose is to give us some perspective on the conclusions some people like to draw from looking at delinquency statistics. Delinquency reports are just a snapshot in time; they are not, themselves, predictions. So the weenies who have been going around trumpeting the fact that, say, 83% of subprime loans are currently current, as if that meant that the 83% will never become delinquent or defaulted in the future, are, well, weenies.

As are some of these opposite-side weenies who seem to think that a 17% delinquency rate means 17% losses to a pool. That would be true only if the "roll rate" of all delinquencies-to-default is 100% and all foreclosed homes have a sales price of $1.00. Not likely, to say the least. Even if the calculations in this example pool are off by a great deal--if the true total defaults end up 30% instead of 20% and the severity is 50% instead of 35%--you get a cumulative loss of 15% of current balance. You can, if you're like me, be convinced that there are some really, really cruddy pools out there that will have numbers at 18 months that look a lot worse than this example from Fitch. I do not doubt that there will also be at least a few that do better. Fitch's assumptions about home price appreciation may be off a little or off the wall: that's one reason we keep watching these trends so closely here at Calculated Risk.

It is particularly annoying when we see people throwing around delinquency numbers as if they were loss percentages, and then applying that number to the total outstanding balance of all subprime securitizations to come up with some scary-looking number in the billions, without taking into account, among other things, the original expected loss on the pools. Not even the meanest (sane) critic of the rating agencies accuses them of having predicted no losses at all when these deals were originally rated. If that were so, there'd be no OC or excess spread and all bonds would be AAA. I think a few folks are getting worked up over the failure of the rating agencies to downgrade some of these pools that have what look like high delinquency rates, without questioning the extent to which those rates are or are not in excess of the original projections. Looking at the example pool above, for instance, you see a projected loss after 18 months, but what you don't see is what the original projection was. Without that, there's no way to conclude that this pool deserves a downgrade.

You might want to plow all the way through the document from which this example is taken if you want to know more about how a delinquency analysis is part of a potential downgrade of a security. It's called "U.S. Subprime RMBS/HEL Upgrade/Downgrade Criteria," and it is available here to registered users of Fitch's website.